Adding upstream version 6.1.76.upstream/6.1.76upstream

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

15 files changed, 1321 insertions, 0 deletions

diff --git a/tools/testing/selftests/powerpc/math/.gitignore b/tools/testing/selftests/powerpc/math/.gitignore
new file mode 100644
index 000000000..07b4893ef
--- /dev/null
+++ b/tools/testing/selftests/powerpc/math/.gitignore
@@ -0,0 +1,10 @@
+# SPDX-License-Identifier: GPL-2.0-only
+fpu_syscall
+vmx_syscall
+fpu_preempt
+vmx_preempt
+fpu_signal
+vmx_signal
+vsx_preempt
+fpu_denormal
+mma
diff --git a/tools/testing/selftests/powerpc/math/Makefile b/tools/testing/selftests/powerpc/math/Makefile
new file mode 100644
index 000000000..3948f7c51
--- /dev/null
+++ b/tools/testing/selftests/powerpc/math/Makefile
@@ -0,0 +1,21 @@
+# SPDX-License-Identifier: GPL-2.0
+TEST_GEN_PROGS := fpu_syscall fpu_preempt fpu_signal fpu_denormal vmx_syscall vmx_preempt vmx_signal vsx_preempt mma
+
+top_srcdir = ../../../../..
+include ../../lib.mk
+
+$(TEST_GEN_PROGS): ../harness.c
+$(TEST_GEN_PROGS): CFLAGS += -O2 -g -pthread -m64 -maltivec
+
+$(OUTPUT)/fpu_syscall: fpu_asm.S
+$(OUTPUT)/fpu_preempt: fpu_asm.S
+$(OUTPUT)/fpu_signal:  fpu_asm.S
+
+$(OUTPUT)/vmx_syscall: vmx_asm.S ../utils.c
+$(OUTPUT)/vmx_preempt: vmx_asm.S ../utils.c
+$(OUTPUT)/vmx_signal: vmx_asm.S ../utils.c
+
+$(OUTPUT)/vsx_preempt: CFLAGS += -mvsx
+$(OUTPUT)/vsx_preempt: vsx_asm.S ../utils.c
+
+$(OUTPUT)/mma: mma.c mma.S ../utils.c
diff --git a/tools/testing/selftests/powerpc/math/fpu_asm.S b/tools/testing/selftests/powerpc/math/fpu_asm.S
new file mode 100644
index 000000000..9dc0c158f
--- /dev/null
+++ b/tools/testing/selftests/powerpc/math/fpu_asm.S
@@ -0,0 +1,131 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Copyright 2015, Cyril Bur, IBM Corp.
+ */
+
+#include "basic_asm.h"
+#include "fpu_asm.h"
+
+FUNC_START(check_fpu)
+	mr r4,r3
+	li	r3,1 # assume a bad result
+	lfd	f0,0(r4)
+	fcmpu	cr1,f0,f14
+	bne	cr1,1f
+	lfd	f0,8(r4)
+	fcmpu	cr1,f0,f15
+	bne	cr1,1f
+	lfd	f0,16(r4)
+	fcmpu	cr1,f0,f16
+	bne	cr1,1f
+	lfd	f0,24(r4)
+	fcmpu	cr1,f0,f17
+	bne	cr1,1f
+	lfd	f0,32(r4)
+	fcmpu	cr1,f0,f18
+	bne	cr1,1f
+	lfd	f0,40(r4)
+	fcmpu	cr1,f0,f19
+	bne	cr1,1f
+	lfd	f0,48(r4)
+	fcmpu	cr1,f0,f20
+	bne	cr1,1f
+	lfd	f0,56(r4)
+	fcmpu	cr1,f0,f21
+	bne	cr1,1f
+	lfd	f0,64(r4)
+	fcmpu	cr1,f0,f22
+	bne	cr1,1f
+	lfd	f0,72(r4)
+	fcmpu	cr1,f0,f23
+	bne	cr1,1f
+	lfd	f0,80(r4)
+	fcmpu	cr1,f0,f24
+	bne	cr1,1f
+	lfd	f0,88(r4)
+	fcmpu	cr1,f0,f25
+	bne	cr1,1f
+	lfd	f0,96(r4)
+	fcmpu	cr1,f0,f26
+	bne	cr1,1f
+	lfd	f0,104(r4)
+	fcmpu	cr1,f0,f27
+	bne	cr1,1f
+	lfd	f0,112(r4)
+	fcmpu	cr1,f0,f28
+	bne	cr1,1f
+	lfd	f0,120(r4)
+	fcmpu	cr1,f0,f29
+	bne	cr1,1f
+	lfd	f0,128(r4)
+	fcmpu	cr1,f0,f30
+	bne	cr1,1f
+	lfd	f0,136(r4)
+	fcmpu	cr1,f0,f31
+	bne	cr1,1f
+	li	r3,0 # Success!!!
+1:	blr
+
+FUNC_START(test_fpu)
+	# r3 holds pointer to where to put the result of fork
+	# r4 holds pointer to the pid
+	# f14-f31 are non volatiles
+	PUSH_BASIC_STACK(256)
+	PUSH_FPU(256)
+	std	r3,STACK_FRAME_PARAM(0)(sp) # Address of darray
+	std r4,STACK_FRAME_PARAM(1)(sp) # Address of pid
+
+	bl load_fpu
+	nop
+	li	r0,__NR_fork
+	sc
+
+	# pass the result of the fork to the caller
+	ld	r9,STACK_FRAME_PARAM(1)(sp)
+	std	r3,0(r9)
+
+	ld r3,STACK_FRAME_PARAM(0)(sp)
+	bl check_fpu
+	nop
+
+	POP_FPU(256)
+	POP_BASIC_STACK(256)
+	blr
+FUNC_END(test_fpu)
+
+# int preempt_fpu(double *darray, int *threads_running, int *running)
+# On starting will (atomically) decrement not_ready as a signal that the FPU
+# has been loaded with darray. Will proceed to check the validity of the FPU
+# registers while running is not zero.
+FUNC_START(preempt_fpu)
+	PUSH_BASIC_STACK(256)
+	PUSH_FPU(256)
+	std r3,STACK_FRAME_PARAM(0)(sp) # double *darray
+	std r4,STACK_FRAME_PARAM(1)(sp) # int *threads_starting
+	std r5,STACK_FRAME_PARAM(2)(sp) # int *running
+
+	bl load_fpu
+	nop
+
+	sync
+	# Atomic DEC
+	ld r3,STACK_FRAME_PARAM(1)(sp)
+1:	lwarx r4,0,r3
+	addi r4,r4,-1
+	stwcx. r4,0,r3
+	bne- 1b
+
+2:	ld r3,STACK_FRAME_PARAM(0)(sp)
+	bl check_fpu
+	nop
+	cmpdi r3,0
+	bne 3f
+	ld r4,STACK_FRAME_PARAM(2)(sp)
+	ld r5,0(r4)
+	cmpwi r5,0
+	bne 2b
+
+3:	POP_FPU(256)
+	POP_BASIC_STACK(256)
+	blr
+FUNC_END(preempt_fpu)
diff --git a/tools/testing/selftests/powerpc/math/fpu_denormal.c b/tools/testing/selftests/powerpc/math/fpu_denormal.c
new file mode 100644
index 000000000..5f96682ab
--- /dev/null
+++ b/tools/testing/selftests/powerpc/math/fpu_denormal.c
@@ -0,0 +1,38 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright IBM Corp. 2020
+ *
+ * This test attempts to cause a FP denormal exception on POWER8 CPUs. Unfortunately
+ * if the denormal handler is not configured or working properly, this can cause a bad
+ * crash in kernel mode when the kernel tries to save FP registers when the process
+ * exits.
+ */
+
+#include <stdio.h>
+#include <string.h>
+
+#include "utils.h"
+
+static int test_denormal_fpu(void)
+{
+	unsigned int m32;
+	unsigned long m64;
+	volatile float f;
+	volatile double d;
+
+	/* try to induce lfs <denormal> ; stfd */
+
+	m32 = 0x00715fcf; /* random denormal */
+	memcpy((float *)&f, &m32, sizeof(f));
+	d = f;
+	memcpy(&m64, (double *)&d, sizeof(d));
+
+	FAIL_IF((long)(m64 != 0x380c57f3c0000000)); /* renormalised value */
+
+	return 0;
+}
+
+int main(int argc, char *argv[])
+{
+	return test_harness(test_denormal_fpu, "fpu_denormal");
+}
diff --git a/tools/testing/selftests/powerpc/math/fpu_preempt.c b/tools/testing/selftests/powerpc/math/fpu_preempt.c
new file mode 100644
index 000000000..3e5b5663d
--- /dev/null
+++ b/tools/testing/selftests/powerpc/math/fpu_preempt.c
@@ -0,0 +1,110 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright 2015, Cyril Bur, IBM Corp.
+ *
+ * This test attempts to see if the FPU registers change across preemption.
+ * Two things should be noted here a) The check_fpu function in asm only checks
+ * the non volatile registers as it is reused from the syscall test b) There is
+ * no way to be sure preemption happened so this test just uses many threads
+ * and a long wait. As such, a successful test doesn't mean much but a failure
+ * is bad.
+ */
+
+#include <stdio.h>
+#include <unistd.h>
+#include <sys/syscall.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <stdlib.h>
+#include <pthread.h>
+
+#include "utils.h"
+
+/* Time to wait for workers to get preempted (seconds) */
+#define PREEMPT_TIME 20
+/*
+ * Factor by which to multiply number of online CPUs for total number of
+ * worker threads
+ */
+#define THREAD_FACTOR 8
+
+
+__thread double darray[] = {0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0,
+		     1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9, 2.0,
+		     2.1};
+
+int threads_starting;
+int running;
+
+extern int preempt_fpu(double *darray, int *threads_starting, int *running);
+
+void *preempt_fpu_c(void *p)
+{
+	long rc;
+	int i;
+
+	srand(pthread_self());
+	for (i = 0; i < 21; i++)
+		darray[i] = rand();
+
+	rc = preempt_fpu(darray, &threads_starting, &running);
+
+	return (void *)rc;
+}
+
+int test_preempt_fpu(void)
+{
+	int i, rc, threads;
+	pthread_t *tids;
+
+	threads = sysconf(_SC_NPROCESSORS_ONLN) * THREAD_FACTOR;
+	tids = malloc((threads) * sizeof(pthread_t));
+	FAIL_IF(!tids);
+
+	running = true;
+	threads_starting = threads;
+	for (i = 0; i < threads; i++) {
+		rc = pthread_create(&tids[i], NULL, preempt_fpu_c, NULL);
+		FAIL_IF(rc);
+	}
+
+	setbuf(stdout, NULL);
+	/* Not really necessary but nice to wait for every thread to start */
+	printf("\tWaiting for all workers to start...");
+	while(threads_starting)
+		asm volatile("": : :"memory");
+	printf("done\n");
+
+	printf("\tWaiting for %d seconds to let some workers get preempted...", PREEMPT_TIME);
+	sleep(PREEMPT_TIME);
+	printf("done\n");
+
+	printf("\tStopping workers...");
+	/*
+	 * Working are checking this value every loop. In preempt_fpu 'cmpwi r5,0; bne 2b'.
+	 * r5 will have loaded the value of running.
+	 */
+	running = 0;
+	for (i = 0; i < threads; i++) {
+		void *rc_p;
+		pthread_join(tids[i], &rc_p);
+
+		/*
+		 * Harness will say the fail was here, look at why preempt_fpu
+		 * returned
+		 */
+		if ((long) rc_p)
+			printf("oops\n");
+		FAIL_IF((long) rc_p);
+	}
+	printf("done\n");
+
+	free(tids);
+	return 0;
+}
+
+int main(int argc, char *argv[])
+{
+	return test_harness(test_preempt_fpu, "fpu_preempt");
+}
diff --git a/tools/testing/selftests/powerpc/math/fpu_signal.c b/tools/testing/selftests/powerpc/math/fpu_signal.c
new file mode 100644
index 000000000..7b1addd50
--- /dev/null
+++ b/tools/testing/selftests/powerpc/math/fpu_signal.c
@@ -0,0 +1,131 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright 2015, Cyril Bur, IBM Corp.
+ *
+ * This test attempts to see if the FPU registers are correctly reported in a
+ * signal context. Each worker just spins checking its FPU registers, at some
+ * point a signal will interrupt it and C code will check the signal context
+ * ensuring it is also the same.
+ */
+
+#include <stdio.h>
+#include <unistd.h>
+#include <sys/syscall.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <stdlib.h>
+#include <pthread.h>
+
+#include "utils.h"
+
+/* Number of times each thread should receive the signal */
+#define ITERATIONS 10
+/*
+ * Factor by which to multiply number of online CPUs for total number of
+ * worker threads
+ */
+#define THREAD_FACTOR 8
+
+__thread double darray[] = {0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0,
+		     1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9, 2.0,
+		     2.1};
+
+bool bad_context;
+int threads_starting;
+int running;
+
+extern long preempt_fpu(double *darray, int *threads_starting, int *running);
+
+void signal_fpu_sig(int sig, siginfo_t *info, void *context)
+{
+	int i;
+	ucontext_t *uc = context;
+	mcontext_t *mc = &uc->uc_mcontext;
+
+	/* Only the non volatiles were loaded up */
+	for (i = 14; i < 32; i++) {
+		if (mc->fp_regs[i] != darray[i - 14]) {
+			bad_context = true;
+			break;
+		}
+	}
+}
+
+void *signal_fpu_c(void *p)
+{
+	int i;
+	long rc;
+	struct sigaction act;
+	act.sa_sigaction = signal_fpu_sig;
+	act.sa_flags = SA_SIGINFO;
+	rc = sigaction(SIGUSR1, &act, NULL);
+	if (rc)
+		return p;
+
+	srand(pthread_self());
+	for (i = 0; i < 21; i++)
+		darray[i] = rand();
+
+	rc = preempt_fpu(darray, &threads_starting, &running);
+
+	return (void *) rc;
+}
+
+int test_signal_fpu(void)
+{
+	int i, j, rc, threads;
+	void *rc_p;
+	pthread_t *tids;
+
+	threads = sysconf(_SC_NPROCESSORS_ONLN) * THREAD_FACTOR;
+	tids = malloc(threads * sizeof(pthread_t));
+	FAIL_IF(!tids);
+
+	running = true;
+	threads_starting = threads;
+	for (i = 0; i < threads; i++) {
+		rc = pthread_create(&tids[i], NULL, signal_fpu_c, NULL);
+		FAIL_IF(rc);
+	}
+
+	setbuf(stdout, NULL);
+	printf("\tWaiting for all workers to start...");
+	while (threads_starting)
+		asm volatile("": : :"memory");
+	printf("done\n");
+
+	printf("\tSending signals to all threads %d times...", ITERATIONS);
+	for (i = 0; i < ITERATIONS; i++) {
+		for (j = 0; j < threads; j++) {
+			pthread_kill(tids[j], SIGUSR1);
+		}
+		sleep(1);
+	}
+	printf("done\n");
+
+	printf("\tStopping workers...");
+	running = 0;
+	for (i = 0; i < threads; i++) {
+		pthread_join(tids[i], &rc_p);
+
+		/*
+		 * Harness will say the fail was here, look at why signal_fpu
+		 * returned
+		 */
+		if ((long) rc_p || bad_context)
+			printf("oops\n");
+		if (bad_context)
+			fprintf(stderr, "\t!! bad_context is true\n");
+		FAIL_IF((long) rc_p || bad_context);
+	}
+	printf("done\n");
+
+	free(tids);
+	return 0;
+}
+
+int main(int argc, char *argv[])
+{
+	return test_harness(test_signal_fpu, "fpu_signal");
+}
diff --git a/tools/testing/selftests/powerpc/math/fpu_syscall.c b/tools/testing/selftests/powerpc/math/fpu_syscall.c
new file mode 100644
index 000000000..694f225c7
--- /dev/null
+++ b/tools/testing/selftests/powerpc/math/fpu_syscall.c
@@ -0,0 +1,86 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright 2015, Cyril Bur, IBM Corp.
+ *
+ * This test attempts to see if the FPU registers change across a syscall (fork).
+ */
+
+#include <stdio.h>
+#include <unistd.h>
+#include <sys/syscall.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <stdlib.h>
+
+#include "utils.h"
+
+extern int test_fpu(double *darray, pid_t *pid);
+
+double darray[] = {0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0,
+		     1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9, 2.0,
+		     2.1};
+
+int syscall_fpu(void)
+{
+	pid_t fork_pid;
+	int i;
+	int ret;
+	int child_ret;
+	for (i = 0; i < 1000; i++) {
+		/* test_fpu will fork() */
+		ret = test_fpu(darray, &fork_pid);
+		if (fork_pid == -1)
+			return -1;
+		if (fork_pid == 0)
+			exit(ret);
+		waitpid(fork_pid, &child_ret, 0);
+		if (ret || child_ret)
+			return 1;
+	}
+
+	return 0;
+}
+
+int test_syscall_fpu(void)
+{
+	/*
+	 * Setup an environment with much context switching
+	 */
+	pid_t pid2;
+	pid_t pid = fork();
+	int ret;
+	int child_ret;
+	FAIL_IF(pid == -1);
+
+	pid2 = fork();
+	/* Can't FAIL_IF(pid2 == -1); because already forked once */
+	if (pid2 == -1) {
+		/*
+		 * Couldn't fork, ensure test is a fail
+		 */
+		child_ret = ret = 1;
+	} else {
+		ret = syscall_fpu();
+		if (pid2)
+			waitpid(pid2, &child_ret, 0);
+		else
+			exit(ret);
+	}
+
+	ret |= child_ret;
+
+	if (pid)
+		waitpid(pid, &child_ret, 0);
+	else
+		exit(ret);
+
+	FAIL_IF(ret || child_ret);
+	return 0;
+}
+
+int main(int argc, char *argv[])
+{
+	return test_harness(test_syscall_fpu, "syscall_fpu");
+
+}
diff --git a/tools/testing/selftests/powerpc/math/mma.S b/tools/testing/selftests/powerpc/math/mma.S
new file mode 100644
index 000000000..61cc88b1b
--- /dev/null
+++ b/tools/testing/selftests/powerpc/math/mma.S
@@ -0,0 +1,36 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later
+ *
+ * Test basic matrix multiply assist (MMA) functionality if available.
+ *
+ * Copyright 2020, Alistair Popple, IBM Corp.
+ */
+	.global test_mma
+test_mma:
+	/* Load accumulator via VSX registers from image passed in r3 */
+	lxvh8x	4,0,3
+	lxvh8x	5,0,4
+
+	/* Clear and prime the accumulator (xxsetaccz) */
+	.long	0x7c030162
+
+	/* Prime the accumulator with MMA VSX move to accumulator
+	* X-form (xxmtacc) (not needed due to above zeroing) */
+	//.long 0x7c010162
+
+	/* xvi16ger2s */
+	.long	0xec042958
+
+	/* Deprime the accumulator - xxmfacc 0 */
+	.long 0x7c000162
+
+	/* Store result in image passed in r5 */
+	stxvw4x	0,0,5
+	addi	5,5,16
+	stxvw4x	1,0,5
+	addi	5,5,16
+	stxvw4x	2,0,5
+	addi	5,5,16
+	stxvw4x	3,0,5
+	addi	5,5,16
+
+	blr
diff --git a/tools/testing/selftests/powerpc/math/mma.c b/tools/testing/selftests/powerpc/math/mma.c
new file mode 100644
index 000000000..3a71808c9
--- /dev/null
+++ b/tools/testing/selftests/powerpc/math/mma.c
@@ -0,0 +1,48 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Test basic matrix multiply assist (MMA) functionality if available.
+ *
+ * Copyright 2020, Alistair Popple, IBM Corp.
+ */
+#include <stdio.h>
+#include <stdint.h>
+
+#include "utils.h"
+
+extern void test_mma(uint16_t (*)[8], uint16_t (*)[8], uint32_t (*)[4*4]);
+
+static int mma(void)
+{
+	int i;
+	int rc = 0;
+	uint16_t x[] = {1, 0, 2, 0, 3, 0, 4, 0};
+	uint16_t y[] = {1, 0, 2, 0, 3, 0, 4, 0};
+	uint32_t z[4*4];
+	uint32_t exp[4*4] = {1, 2, 3, 4,
+			     2, 4, 6, 8,
+			     3, 6, 9, 12,
+			     4, 8, 12, 16};
+
+	SKIP_IF_MSG(!have_hwcap2(PPC_FEATURE2_ARCH_3_1), "Need ISAv3.1");
+	SKIP_IF_MSG(!have_hwcap2(PPC_FEATURE2_MMA), "Need MMA");
+
+	test_mma(&x, &y, &z);
+
+	for (i = 0; i < 16; i++) {
+		printf("MMA[%d] = %d ", i, z[i]);
+
+		if (z[i] == exp[i]) {
+			printf(" (Correct)\n");
+		} else {
+			printf(" (Incorrect)\n");
+			rc = 1;
+		}
+	}
+
+	return rc;
+}
+
+int main(int argc, char *argv[])
+{
+	return test_harness(mma, "mma");
+}
diff --git a/tools/testing/selftests/powerpc/math/vmx_asm.S b/tools/testing/selftests/powerpc/math/vmx_asm.S
new file mode 100644
index 000000000..11b0704c5
--- /dev/null
+++ b/tools/testing/selftests/powerpc/math/vmx_asm.S
@@ -0,0 +1,148 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Copyright 2015, Cyril Bur, IBM Corp.
+ */
+
+#include "basic_asm.h"
+#include "vmx_asm.h"
+
+# Should be safe from C, only touches r4, r5 and v0,v1,v2
+FUNC_START(check_vmx)
+	PUSH_BASIC_STACK(32)
+	mr r4,r3
+	li	r3,1 # assume a bad result
+	li	r5,0
+	lvx	v0,r5,r4
+	vcmpequd.	v1,v0,v20
+	vmr	v2,v1
+
+	addi	r5,r5,16
+	lvx	v0,r5,r4
+	vcmpequd.	v1,v0,v21
+	vand	v2,v2,v1
+
+	addi	r5,r5,16
+	lvx	v0,r5,r4
+	vcmpequd.	v1,v0,v22
+	vand	v2,v2,v1
+
+	addi	r5,r5,16
+	lvx	v0,r5,r4
+	vcmpequd.	v1,v0,v23
+	vand	v2,v2,v1
+
+	addi	r5,r5,16
+	lvx	v0,r5,r4
+	vcmpequd.	v1,v0,v24
+	vand	v2,v2,v1
+
+	addi	r5,r5,16
+	lvx	v0,r5,r4
+	vcmpequd.	v1,v0,v25
+	vand	v2,v2,v1
+
+	addi	r5,r5,16
+	lvx	v0,r5,r4
+	vcmpequd.	v1,v0,v26
+	vand	v2,v2,v1
+
+	addi	r5,r5,16
+	lvx	v0,r5,r4
+	vcmpequd.	v1,v0,v27
+	vand	v2,v2,v1
+
+	addi	r5,r5,16
+	lvx	v0,r5,r4
+	vcmpequd.	v1,v0,v28
+	vand	v2,v2,v1
+
+	addi	r5,r5,16
+	lvx	v0,r5,r4
+	vcmpequd.	v1,v0,v29
+	vand	v2,v2,v1
+
+	addi	r5,r5,16
+	lvx	v0,r5,r4
+	vcmpequd.	v1,v0,v30
+	vand	v2,v2,v1
+
+	addi	r5,r5,16
+	lvx	v0,r5,r4
+	vcmpequd.	v1,v0,v31
+	vand	v2,v2,v1
+
+	li	r5,STACK_FRAME_LOCAL(0,0)
+	stvx	v2,r5,sp
+	ldx	r0,r5,sp
+	cmpdi	r0,0xffffffffffffffff
+	bne	1f
+	li	r3,0
+1:	POP_BASIC_STACK(32)
+	blr
+FUNC_END(check_vmx)
+
+# Safe from C
+FUNC_START(test_vmx)
+	# r3 holds pointer to where to put the result of fork
+	# r4 holds pointer to the pid
+	# v20-v31 are non-volatile
+	PUSH_BASIC_STACK(512)
+	std	r3,STACK_FRAME_PARAM(0)(sp) # Address of varray
+	std r4,STACK_FRAME_PARAM(1)(sp) # address of pid
+	PUSH_VMX(STACK_FRAME_LOCAL(2,0),r4)
+
+	bl load_vmx
+	nop
+
+	li	r0,__NR_fork
+	sc
+	# Pass the result of fork back to the caller
+	ld	r9,STACK_FRAME_PARAM(1)(sp)
+	std	r3,0(r9)
+
+	ld r3,STACK_FRAME_PARAM(0)(sp)
+	bl check_vmx
+	nop
+
+	POP_VMX(STACK_FRAME_LOCAL(2,0),r4)
+	POP_BASIC_STACK(512)
+	blr
+FUNC_END(test_vmx)
+
+# int preempt_vmx(vector int *varray, int *threads_starting, int *running)
+# On starting will (atomically) decrement threads_starting as a signal that
+# the VMX have been loaded with varray. Will proceed to check the validity of
+# the VMX registers while running is not zero.
+FUNC_START(preempt_vmx)
+	PUSH_BASIC_STACK(512)
+	std r3,STACK_FRAME_PARAM(0)(sp) # vector int *varray
+	std r4,STACK_FRAME_PARAM(1)(sp) # int *threads_starting
+	std r5,STACK_FRAME_PARAM(2)(sp) # int *running
+	# VMX need to write to 16 byte aligned addresses, skip STACK_FRAME_LOCAL(3,0)
+	PUSH_VMX(STACK_FRAME_LOCAL(4,0),r4)
+
+	bl load_vmx
+	nop
+
+	sync
+	# Atomic DEC
+	ld r3,STACK_FRAME_PARAM(1)(sp)
+1:	lwarx r4,0,r3
+	addi r4,r4,-1
+	stwcx. r4,0,r3
+	bne- 1b
+
+2:	ld r3,STACK_FRAME_PARAM(0)(sp)
+	bl check_vmx
+	nop
+	cmpdi r3,0
+	bne 3f
+	ld r4,STACK_FRAME_PARAM(2)(sp)
+	ld r5,0(r4)
+	cmpwi r5,0
+	bne 2b
+
+3:	POP_VMX(STACK_FRAME_LOCAL(4,0),r4)
+	POP_BASIC_STACK(512)
+	blr
+FUNC_END(preempt_vmx)
diff --git a/tools/testing/selftests/powerpc/math/vmx_preempt.c b/tools/testing/selftests/powerpc/math/vmx_preempt.c
new file mode 100644
index 000000000..6f7cf400c
--- /dev/null
+++ b/tools/testing/selftests/powerpc/math/vmx_preempt.c
@@ -0,0 +1,113 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright 2015, Cyril Bur, IBM Corp.
+ *
+ * This test attempts to see if the VMX registers change across preemption.
+ * Two things should be noted here a) The check_vmx function in asm only checks
+ * the non volatile registers as it is reused from the syscall test b) There is
+ * no way to be sure preemption happened so this test just uses many threads
+ * and a long wait. As such, a successful test doesn't mean much but a failure
+ * is bad.
+ */
+
+#include <stdio.h>
+#include <unistd.h>
+#include <sys/syscall.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <stdlib.h>
+#include <pthread.h>
+
+#include "utils.h"
+
+/* Time to wait for workers to get preempted (seconds) */
+#define PREEMPT_TIME 20
+/*
+ * Factor by which to multiply number of online CPUs for total number of
+ * worker threads
+ */
+#define THREAD_FACTOR 8
+
+__thread vector int varray[] = {{1, 2, 3, 4}, {5, 6, 7, 8}, {9, 10,11,12},
+	{13,14,15,16},{17,18,19,20},{21,22,23,24},
+	{25,26,27,28},{29,30,31,32},{33,34,35,36},
+	{37,38,39,40},{41,42,43,44},{45,46,47,48}};
+
+int threads_starting;
+int running;
+
+extern int preempt_vmx(vector int *varray, int *threads_starting, int *running);
+
+void *preempt_vmx_c(void *p)
+{
+	int i, j;
+	long rc;
+
+	srand(pthread_self());
+	for (i = 0; i < 12; i++)
+		for (j = 0; j < 4; j++)
+			varray[i][j] = rand();
+
+	rc = preempt_vmx(varray, &threads_starting, &running);
+
+	return (void *)rc;
+}
+
+int test_preempt_vmx(void)
+{
+	int i, rc, threads;
+	pthread_t *tids;
+
+	// vcmpequd used in vmx_asm.S is v2.07
+	SKIP_IF(!have_hwcap2(PPC_FEATURE2_ARCH_2_07));
+
+	threads = sysconf(_SC_NPROCESSORS_ONLN) * THREAD_FACTOR;
+	tids = malloc(threads * sizeof(pthread_t));
+	FAIL_IF(!tids);
+
+	running = true;
+	threads_starting = threads;
+	for (i = 0; i < threads; i++) {
+		rc = pthread_create(&tids[i], NULL, preempt_vmx_c, NULL);
+		FAIL_IF(rc);
+	}
+
+	setbuf(stdout, NULL);
+	/* Not really nessesary but nice to wait for every thread to start */
+	printf("\tWaiting for all workers to start...");
+	while(threads_starting)
+		asm volatile("": : :"memory");
+	printf("done\n");
+
+	printf("\tWaiting for %d seconds to let some workers get preempted...", PREEMPT_TIME);
+	sleep(PREEMPT_TIME);
+	printf("done\n");
+
+	printf("\tStopping workers...");
+	/*
+	 * Working are checking this value every loop. In preempt_vmx 'cmpwi r5,0; bne 2b'.
+	 * r5 will have loaded the value of running.
+	 */
+	running = 0;
+	for (i = 0; i < threads; i++) {
+		void *rc_p;
+		pthread_join(tids[i], &rc_p);
+
+		/*
+		 * Harness will say the fail was here, look at why preempt_vmx
+		 * returned
+		 */
+		if ((long) rc_p)
+			printf("oops\n");
+		FAIL_IF((long) rc_p);
+	}
+	printf("done\n");
+
+	return 0;
+}
+
+int main(int argc, char *argv[])
+{
+	return test_harness(test_preempt_vmx, "vmx_preempt");
+}
diff --git a/tools/testing/selftests/powerpc/math/vmx_signal.c b/tools/testing/selftests/powerpc/math/vmx_signal.c
new file mode 100644
index 000000000..b340a5c4e
--- /dev/null
+++ b/tools/testing/selftests/powerpc/math/vmx_signal.c
@@ -0,0 +1,155 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright 2015, Cyril Bur, IBM Corp.
+ *
+ * This test attempts to see if the VMX registers are correctly reported in a
+ * signal context. Each worker just spins checking its VMX registers, at some
+ * point a signal will interrupt it and C code will check the signal context
+ * ensuring it is also the same.
+ */
+
+#include <stdio.h>
+#include <unistd.h>
+#include <sys/syscall.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <stdlib.h>
+#include <string.h>
+#include <pthread.h>
+#include <altivec.h>
+
+#include "utils.h"
+
+/* Number of times each thread should receive the signal */
+#define ITERATIONS 10
+/*
+ * Factor by which to multiply number of online CPUs for total number of
+ * worker threads
+ */
+#define THREAD_FACTOR 8
+
+__thread vector int varray[] = {{1, 2, 3, 4}, {5, 6, 7, 8}, {9, 10,11,12},
+	{13,14,15,16},{17,18,19,20},{21,22,23,24},
+	{25,26,27,28},{29,30,31,32},{33,34,35,36},
+	{37,38,39,40},{41,42,43,44},{45,46,47,48}};
+
+bool bad_context;
+int running;
+int threads_starting;
+
+extern int preempt_vmx(vector int *varray, int *threads_starting, int *sentinal);
+
+void signal_vmx_sig(int sig, siginfo_t *info, void *context)
+{
+	int i;
+	ucontext_t *uc = context;
+	mcontext_t *mc = &uc->uc_mcontext;
+
+	/* Only the non volatiles were loaded up */
+	for (i = 20; i < 32; i++) {
+		if (memcmp(mc->v_regs->vrregs[i], &varray[i - 20], 16)) {
+			int j;
+			/*
+			 * Shouldn't printf() in a signal handler, however, this is a
+			 * test and we've detected failure. Understanding what failed
+			 * is paramount. All that happens after this is tests exit with
+			 * failure.
+			 */
+			printf("VMX mismatch at reg %d!\n", i);
+			printf("Reg | Actual                  | Expected\n");
+			for (j = 20; j < 32; j++) {
+				printf("%d  | 0x%04x%04x%04x%04x      | 0x%04x%04x%04x%04x\n", j, mc->v_regs->vrregs[j][0],
+					   mc->v_regs->vrregs[j][1], mc->v_regs->vrregs[j][2], mc->v_regs->vrregs[j][3],
+					   varray[j - 20][0], varray[j - 20][1], varray[j - 20][2], varray[j - 20][3]);
+			}
+			bad_context = true;
+			break;
+		}
+	}
+}
+
+void *signal_vmx_c(void *p)
+{
+	int i, j;
+	long rc;
+	struct sigaction act;
+	act.sa_sigaction = signal_vmx_sig;
+	act.sa_flags = SA_SIGINFO;
+	rc = sigaction(SIGUSR1, &act, NULL);
+	if (rc)
+		return p;
+
+	srand(pthread_self());
+	for (i = 0; i < 12; i++)
+		for (j = 0; j < 4; j++)
+			varray[i][j] = rand();
+
+	rc = preempt_vmx(varray, &threads_starting, &running);
+
+	return (void *) rc;
+}
+
+int test_signal_vmx(void)
+{
+	int i, j, rc, threads;
+	void *rc_p;
+	pthread_t *tids;
+
+	// vcmpequd used in vmx_asm.S is v2.07
+	SKIP_IF(!have_hwcap2(PPC_FEATURE2_ARCH_2_07));
+
+	threads = sysconf(_SC_NPROCESSORS_ONLN) * THREAD_FACTOR;
+	tids = malloc(threads * sizeof(pthread_t));
+	FAIL_IF(!tids);
+
+	running = true;
+	threads_starting = threads;
+	for (i = 0; i < threads; i++) {
+		rc = pthread_create(&tids[i], NULL, signal_vmx_c, NULL);
+		FAIL_IF(rc);
+	}
+
+	setbuf(stdout, NULL);
+	printf("\tWaiting for %d workers to start... %d", threads, threads_starting);
+	while (threads_starting) {
+		asm volatile("": : :"memory");
+		usleep(1000);
+		printf(", %d", threads_starting);
+	}
+	printf(" ...done\n");
+
+	printf("\tSending signals to all threads %d times...", ITERATIONS);
+	for (i = 0; i < ITERATIONS; i++) {
+		for (j = 0; j < threads; j++) {
+			pthread_kill(tids[j], SIGUSR1);
+		}
+		sleep(1);
+	}
+	printf("done\n");
+
+	printf("\tKilling workers...");
+	running = 0;
+	for (i = 0; i < threads; i++) {
+		pthread_join(tids[i], &rc_p);
+
+		/*
+		 * Harness will say the fail was here, look at why signal_vmx
+		 * returned
+		 */
+		if ((long) rc_p || bad_context)
+			printf("oops\n");
+		if (bad_context)
+			fprintf(stderr, "\t!! bad_context is true\n");
+		FAIL_IF((long) rc_p || bad_context);
+	}
+	printf("done\n");
+
+	free(tids);
+	return 0;
+}
+
+int main(int argc, char *argv[])
+{
+	return test_harness(test_signal_vmx, "vmx_signal");
+}
diff --git a/tools/testing/selftests/powerpc/math/vmx_syscall.c b/tools/testing/selftests/powerpc/math/vmx_syscall.c
new file mode 100644
index 000000000..03c78dfe3
--- /dev/null
+++ b/tools/testing/selftests/powerpc/math/vmx_syscall.c
@@ -0,0 +1,92 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright 2015, Cyril Bur, IBM Corp.
+ *
+ * This test attempts to see if the VMX registers change across a syscall (fork).
+ */
+
+#include <altivec.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <sys/syscall.h>
+#include <sys/time.h>
+#include <stdlib.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include "utils.h"
+
+vector int varray[] = {{1, 2, 3, 4}, {5, 6, 7, 8}, {9, 10,11,12},
+	{13,14,15,16},{17,18,19,20},{21,22,23,24},
+	{25,26,27,28},{29,30,31,32},{33,34,35,36},
+	{37,38,39,40},{41,42,43,44},{45,46,47,48}};
+
+extern int test_vmx(vector int *varray, pid_t *pid);
+
+int vmx_syscall(void)
+{
+	pid_t fork_pid;
+	int i;
+	int ret;
+	int child_ret;
+	for (i = 0; i < 1000; i++) {
+		/* test_vmx will fork() */
+		ret = test_vmx(varray, &fork_pid);
+		if (fork_pid == -1)
+			return -1;
+		if (fork_pid == 0)
+			exit(ret);
+		waitpid(fork_pid, &child_ret, 0);
+		if (ret || child_ret)
+			return 1;
+	}
+
+	return 0;
+}
+
+int test_vmx_syscall(void)
+{
+	/*
+	 * Setup an environment with much context switching
+	 */
+	pid_t pid2;
+	pid_t pid;
+	int ret;
+	int child_ret;
+
+	// vcmpequd used in vmx_asm.S is v2.07
+	SKIP_IF(!have_hwcap2(PPC_FEATURE2_ARCH_2_07));
+
+	pid = fork();
+	FAIL_IF(pid == -1);
+
+	pid2 = fork();
+	ret = vmx_syscall();
+	/* Can't FAIL_IF(pid2 == -1); because we've already forked */
+	if (pid2 == -1) {
+		/*
+		 * Couldn't fork, ensure child_ret is set and is a fail
+		 */
+		ret = child_ret = 1;
+	} else {
+		if (pid2)
+			waitpid(pid2, &child_ret, 0);
+		else
+			exit(ret);
+	}
+
+	ret |= child_ret;
+
+	if (pid)
+		waitpid(pid, &child_ret, 0);
+	else
+		exit(ret);
+
+	FAIL_IF(ret || child_ret);
+	return 0;
+}
+
+int main(int argc, char *argv[])
+{
+	return test_harness(test_vmx_syscall, "vmx_syscall");
+
+}
diff --git a/tools/testing/selftests/powerpc/math/vsx_asm.S b/tools/testing/selftests/powerpc/math/vsx_asm.S
new file mode 100644
index 000000000..ffc165d98
--- /dev/null
+++ b/tools/testing/selftests/powerpc/math/vsx_asm.S
@@ -0,0 +1,57 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Copyright 2015, Cyril Bur, IBM Corp.
+ */
+
+#include "basic_asm.h"
+#include "vsx_asm.h"
+
+#long check_vsx(vector int *r3);
+#This function wraps storeing VSX regs to the end of an array and a
+#call to a comparison function in C which boils down to a memcmp()
+FUNC_START(check_vsx)
+	PUSH_BASIC_STACK(32)
+	std	r3,STACK_FRAME_PARAM(0)(sp)
+	addi r3, r3, 16 * 12 #Second half of array
+	bl store_vsx
+	ld r3,STACK_FRAME_PARAM(0)(sp)
+	bl vsx_memcmp
+	POP_BASIC_STACK(32)
+	blr
+FUNC_END(check_vsx)
+
+# int preempt_vmx(vector int *varray, int *threads_starting,
+#                 int *running);
+# On starting will (atomically) decrement threads_starting as a signal
+# that the VMX have been loaded with varray. Will proceed to check the
+# validity of the VMX registers while running is not zero.
+FUNC_START(preempt_vsx)
+	PUSH_BASIC_STACK(512)
+	std r3,STACK_FRAME_PARAM(0)(sp) # vector int *varray
+	std r4,STACK_FRAME_PARAM(1)(sp) # int *threads_starting
+	std r5,STACK_FRAME_PARAM(2)(sp) # int *running
+
+	bl load_vsx
+	nop
+
+	sync
+	# Atomic DEC
+	ld r3,STACK_FRAME_PARAM(1)(sp)
+1:	lwarx r4,0,r3
+	addi r4,r4,-1
+	stwcx. r4,0,r3
+	bne- 1b
+
+2:	ld r3,STACK_FRAME_PARAM(0)(sp)
+	bl check_vsx
+	nop
+	cmpdi r3,0
+	bne 3f
+	ld r4,STACK_FRAME_PARAM(2)(sp)
+	ld r5,0(r4)
+	cmpwi r5,0
+	bne 2b
+
+3:	POP_BASIC_STACK(512)
+	blr
+FUNC_END(preempt_vsx)
diff --git a/tools/testing/selftests/powerpc/math/vsx_preempt.c b/tools/testing/selftests/powerpc/math/vsx_preempt.c
new file mode 100644
index 000000000..d1601bb88
--- /dev/null
+++ b/tools/testing/selftests/powerpc/math/vsx_preempt.c
@@ -0,0 +1,145 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright 2015, Cyril Bur, IBM Corp.
+ *
+ * This test attempts to see if the VSX registers change across preemption.
+ * There is no way to be sure preemption happened so this test just
+ * uses many threads and a long wait. As such, a successful test
+ * doesn't mean much but a failure is bad.
+ */
+
+#include <stdio.h>
+#include <string.h>
+#include <unistd.h>
+#include <sys/syscall.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <stdlib.h>
+#include <pthread.h>
+
+#include "utils.h"
+
+/* Time to wait for workers to get preempted (seconds) */
+#define PREEMPT_TIME 20
+/*
+ * Factor by which to multiply number of online CPUs for total number of
+ * worker threads
+ */
+#define THREAD_FACTOR 8
+
+/*
+ * Ensure there is twice the number of non-volatile VMX regs!
+ * check_vmx() is going to use the other half as space to put the live
+ * registers before calling vsx_memcmp()
+ */
+__thread vector int varray[24] = {
+	{1, 2, 3, 4 }, {5, 6, 7, 8 }, {9, 10,11,12},
+	{13,14,15,16}, {17,18,19,20}, {21,22,23,24},
+	{25,26,27,28}, {29,30,31,32}, {33,34,35,36},
+	{37,38,39,40}, {41,42,43,44}, {45,46,47,48}
+};
+
+int threads_starting;
+int running;
+
+extern long preempt_vsx(vector int *varray, int *threads_starting, int *running);
+
+long vsx_memcmp(vector int *a) {
+	vector int zero = {0, 0, 0, 0};
+	int i;
+
+	FAIL_IF(a != varray);
+
+	for(i = 0; i < 12; i++) {
+		if (memcmp(&a[i + 12], &zero, sizeof(vector int)) == 0) {
+			fprintf(stderr, "Detected zero from the VSX reg %d\n", i + 12);
+			return 2;
+		}
+	}
+
+	if (memcmp(a, &a[12], 12 * sizeof(vector int))) {
+		long *p = (long *)a;
+		fprintf(stderr, "VSX mismatch\n");
+		for (i = 0; i < 24; i=i+2)
+			fprintf(stderr, "%d: 0x%08lx%08lx | 0x%08lx%08lx\n",
+					i/2 + i%2 + 20, p[i], p[i + 1], p[i + 24], p[i + 25]);
+		return 1;
+	}
+	return 0;
+}
+
+void *preempt_vsx_c(void *p)
+{
+	int i, j;
+	long rc;
+	srand(pthread_self());
+	for (i = 0; i < 12; i++)
+		for (j = 0; j < 4; j++) {
+			varray[i][j] = rand();
+			/* Don't want zero because it hides kernel problems */
+			if (varray[i][j] == 0)
+				j--;
+		}
+	rc = preempt_vsx(varray, &threads_starting, &running);
+	if (rc == 2)
+		fprintf(stderr, "Caught zeros in VSX compares\n");
+	return (void *)rc;
+}
+
+int test_preempt_vsx(void)
+{
+	int i, rc, threads;
+	pthread_t *tids;
+
+	SKIP_IF(!have_hwcap(PPC_FEATURE_HAS_VSX));
+
+	threads = sysconf(_SC_NPROCESSORS_ONLN) * THREAD_FACTOR;
+	tids = malloc(threads * sizeof(pthread_t));
+	FAIL_IF(!tids);
+
+	running = true;
+	threads_starting = threads;
+	for (i = 0; i < threads; i++) {
+		rc = pthread_create(&tids[i], NULL, preempt_vsx_c, NULL);
+		FAIL_IF(rc);
+	}
+
+	setbuf(stdout, NULL);
+	/* Not really nessesary but nice to wait for every thread to start */
+	printf("\tWaiting for %d workers to start...", threads_starting);
+	while(threads_starting)
+		asm volatile("": : :"memory");
+	printf("done\n");
+
+	printf("\tWaiting for %d seconds to let some workers get preempted...", PREEMPT_TIME);
+	sleep(PREEMPT_TIME);
+	printf("done\n");
+
+	printf("\tStopping workers...");
+	/*
+	 * Working are checking this value every loop. In preempt_vsx 'cmpwi r5,0; bne 2b'.
+	 * r5 will have loaded the value of running.
+	 */
+	running = 0;
+	for (i = 0; i < threads; i++) {
+		void *rc_p;
+		pthread_join(tids[i], &rc_p);
+
+		/*
+		 * Harness will say the fail was here, look at why preempt_vsx
+		 * returned
+		 */
+		if ((long) rc_p)
+			printf("oops\n");
+		FAIL_IF((long) rc_p);
+	}
+	printf("done\n");
+
+	return 0;
+}
+
+int main(int argc, char *argv[])
+{
+	return test_harness(test_preempt_vsx, "vsx_preempt");
+}